Introducing NVIDIA Tesla P4 GPUs for accelerating virtual workstations and ML inference on Compute Engine
Product Manager, Google Compute Engine
Today, we are excited to announce a new addition to the Google Cloud Platform (GCP) GPU family that’s optimized for graphics-intensive applications and machine learning inference: the NVIDIA Tesla P4 GPU.
We’ve come a long way since we introduced our first-generation compute accelerator, the K80 GPU, adding along the way P100 and V100 GPUs that are optimized for machine learning and HPC workloads. The new P4 accelerators, now in beta, provide a good balance of price/performance for remote display applications and real-time machine learning inference.
Graphics-intensive applications that run in the cloud benefit greatly from workstation-class GPUs. We now support virtual workstations with NVIDIA GRID on the P4 and P100, allowing you to turn any instance with one or more GPUs into a high-end workstation optimized for graphics-accelerated use cases. Now, artists, architects and engineers can create breathtaking 3D scenes for their next blockbuster film, or design a computer-aided photorealistic composition. P4s offer 8GB of GDDR5 memory— enough power for highly demanding interactive and immersive 3D applications.
One early customer, Schlumberger, has been impressed with the flexibility that Google Cloud virtual workstations offer oil and gas customers.
“Compute Engine Virtual Workstations powered by NVIDIA GPUs allow us to address our cloud-based users' largest visualization challenges. Oil and Gas subject-matter experts can now use the Petrotechnical Suite in the DELFI cognitive E&P environment from Schlumberger on various devices with an expanding global reach. With Google Cloud’s network and increasing GPU capacity to deliver low-latency, virtual workstation experiences, our customers are able to leverage the power of cloud computing to make more informed decisions from oil and gas exploration to production.” - Raj Kannan, Advisor Solution Architect, Schlumberger
Using GPUs for cloud-based 3D applications is a must. To help make the best possible virtual workstations, we’ve also partnered with Teradici, which offers a client that runs on Compute Engine and that is available through GCP Marketplace.
"The combination of Teradici Cloud Access Software, Google Cloud Platform and NVIDIA P4 and P100 instances helps deliver high-performance virtual workstation experiences. Now customers in industries such as media and entertainment, oil & gas and manufacturing, can empower their end users to run compute-intensive tools from nearly any device, virtually anywhere in the world." - Dan Cordingley, CEO, Teradici
NVIDIA Tesla P4 GPUs are also a great fit for ML inference use cases such as visual search, interactive speech and video recommendations. These accelerators offer up to 22 TOPs of INT8 performance and can slash latency by 40X compared to traditional CPUs. Finally, P4s are a good fit for video transcoding workloads. The P4’s one Decode and two Encode engines can transcode up to 35 HD video streams in real-time 1, 2.
Like with our other GPU offerings, you can attach one or multiple P4s to any machine type, and pay only for the resources that you need. We support all our GPUs, including the P4, on Kubernetes Engine and Cloud Machine Learning Engine, zync render, and they can take advantage of sustained use discounts and preemptible pricing. P4s are available today in select zones in us-central1 (Iowa), us-east4 (N. Virginia), Montreal (northamerica-northeast1) and europe-west4 (Netherlands), with more regions (including LA), coming soon. The below table summarizes our full set of GPU offerings.
Our goal is to collaborate with you and provide the hardware resources, like accelerators, to power all of your workloads—whether it is machine learning inference and training, simulation, genomics, HPC, rendering and now, 3D visualization. To learn more about P4 and virtual workstations on GCP, check out the GPU or NVIDIA page. Then, visit the console and get started!
* Maximum vCPU count and system memory limit on the instance might be smaller depending on the zone or the number of GPUs selected. ** GPU prices listed as hourly rate, per GPU attached to a VM that are billed by the second. Pricing for attaching GPUs to preemptible VMs is different from pricing for attaching GPUs to non-preemptible VMs. Starting prices listed are for some US regions. Prices for other regions may be different. Additional Sustained Use Discounts of up to 30% apply to GPU non-preemptible usage only.